A.I - Deep Learning & Machine Learning: Education and Discussion

Software & Operating Systems
,
1

I decided to create a thread containing information regarding the recent advent of machine learning and deep learning. This thread is also for the discussion of ML and DL, implementations and ethics.

What is Machine Learning?

Ok, but how does it work?

Start here:

I don't want to take a course but I would like to have a technical read

What is Deep Learning?

Ok, but how does it work?

Start here:
http://course.fast.ai/start.html

I don't want to take a course but I would like to have a technical read

http://www.deeplearningbook.org/

I would reccomend taking the courses above but have a read if you want. The following sub reddits are very good:

https://www.reddit.com/r/deeplearning/

Heres a awesome list:

3 Likes
2

For anyone who wants to get into Machine learning I have read and can highly recommend these two books:

This one gives a full introduction int ML concepts, including preparing the data, selecting models, optimizions, evaluation, models and ensembling multiple classifiers to boost accuracy. Excellent book for newbies.

Even if you don’t want to buy the book all jupyter notebooks are available for free: https://github.com/ageron/handson-ml

The second one assumes you to be somewhat familiar with ML concepts and only focuses on Keras. Keras is a python library that makes writing neural networks a breeze. Some of the best image classification networks in existence can be written in less than 40 lines of code.

This one has some code available as well, including large sections of the book: https://github.com/fchollet/deep-learning-with-python-notebooks

EDIT: Example network for classifying natural language movie reviews into positive/negative:

model = models.Sequential()
model.add(layers.Dense(16, activation='relu', input_shape=(10000,)))
model.add(layers.Dense(16, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))
model.compile(optimizer='rmsprop',
              loss='binary_crossentropy',
              metrics=['accuracy'])

That’s the entire network. Keras ftw.

5 Likes
3

I might have to give keras a look

4

Have you had a look at fast.ai? There library seems to be the most competitive imo.

1 Like
5

Not yet. I hear people really like it so I’ll give it a try once I’m done exploring Keras.

1 Like
6

Although I realise many of the ML applications are generally run in Python, I’ve run a fair bit with R in R-Studio. I’ve used the MLR package:

https://mlr-org.github.io/mlr-tutorial/devel/html/

I will say though that much of the ML stuff in R is based on data-analytics as opposed to image recognition etc. It’s a really good alternative from the traditional statistical techniques for analysing huge data-sets (i.e. greater than 10k - although I’ve had varying results with around 3k as well).

As for predictive text-books, I’d recommend this one:

Absolutely awesome book that give’s a great explanation for supervised machine learning. Great starting point for anyone interested in ML for data-analytics. It has a great section on data-wrangling and data-manipulation as well (Don’t forget to clean your data, folks!).

7

This points to what I think is the biggest misconception in ML right now. Using estimators and creating neural networks is the easy part. The real challenge is preparing the data and extracting meaningful features.

3 Likes
8

I’ll always remember my lecturer’s words…

“80-20 rule. 80% pre-processing, 20% everything else!”

2 Likes
9

Lots of clean data.

I did find this interesting. From Linux Unplugged

To break the WWII Enigma enginge. They trained an AI with german childrens books to get a baseline AI. Then spun up 1000 then later 2000 indepenant AI’s running to break the code and report back. Its an interesting read.

I bit like 1000 Monkey’s er AI’s with typewriters took 1000 years er 13 mins.

Most interesting was the cost to do this. Was under $20 of processing so they say.

10

image

ok